A computational econometric modeling technique for the parametric simulation of effort and duration estimation as applied to the development of software systems.

摘要

Software development continues to be challenged by projects that are often late or over budget. In this dissertation we review the relevant traditional and agile approaches to the governance control and management of software projects. We explore reasons for common disconnects between initial estimates and the final resulting cost, duration and effort metrics for software projects. We propose that improving the accuracy of initial estimation would result in a better match between expectations and results in software development. Finally we introduce a detailed parametric modeling and simulation approach for developing early effort and schedule estimates for a software project.;The detailed parametric model for software project effort and schedule estimation proposed here incorporates granular parametric modeling techniques coupled with process simulation and analysis techniques in order to address a broader range of project scenarios. We propose that simpler versions of the parametric basis formulas can be employed and that process analysis can improve the utility of the effort and schedule estimation for real world projects thus improving the management of risk. The model also incorporates a re-work feedback loop that models the disruptive behavior of bugs. We propose that the 'transfer functions' used in this model provide a more granular approach that can be tuned locally for an on-going project as work begins and provides more actionable and immediate feedback. We further propose that this detailed parametric model deals with real world problems such as changing requirements and variations in resource availability in ways that traditional parametric models cannot.;While the model developed here is based on a specific software development life cycle (Unified Process) used in the case study project for a Fortune 500 company, it can be applied to any defined methodology that can be represented by stages of work. These software development life cycle stages of work can be linked sequentially or in parallel to complete a given final work product or products. In fact, the modeling and simulation methodology developed here can be applied to any business process life cycle, not just software development.